Enhancing Robotic Perception and Adaptability through Sensor Fusion and Origami-Inspired Designs

Namai Chandra; Jaison Jose; Kavi Arya; Shivaram Kalyanakrishnan

cs.ROOct 7, 2026

Enhancing Robotic Perception and Adaptability through Sensor Fusion and Origami-Inspired Designs

Authors: Namai Chandra, Jaison Jose, Kavi Arya, Shivaram Kalyanakrishnan

Organizations: IIT Madras · e-Yantra, IIT Bombay

Abstract

Compact mobile robots must recover scene geometry under changing lighting and surface texture while working within tight payload and cost limits. We present a compact mobile robot that uses origami-inspired wheels for locomotion and active control of its sensing geometry. As the wheels move between terrain-adaptive configurations, the changing chassis pitch sweeps a 2D LiDAR through intermediate elevations; held wheel positions provide a chosen viewing angle. An IMU accounts for chassis attitude, and a fusion node projects LiDAR returns into the RGB-D depth stream supplied to RTAB-Map. The arrangement uses the wheel actuation already present on a sub-300 USD, sub-2 kg prototype to extend the scanner's viewing geometry. We assess depth fusion in a textureless indoor corridor and an outdoor sunlit area, with three runs per sensor configuration in each setting. Mean full-frame invalid-depth fractions fell from 21% to 11% indoors and from 48% to 18% outdoors. The prototype combines improved depth coverage with a continuously adjustable LiDAR viewpoint using the same actuation that reconfigures its wheels.

Figures & tables

Panel of Figure 1 — Fig. 1: System overview. Left: sensor suite combining a RealSense D435i RGB-D camera, an RPLiDAR A2M12 2D scanner, and a BNO055 IMU. Center (Contribution 1): confidence-weighted fusion pipeline that blends LiDAR depth into the RGB-D stream before passing the result to RTAB-Map. Right (Contribution 2): origami morphing wheels tilt the chassis to sweep the 2D LiDAR vertically; the IMU compensates the tilt in software. The flat (left) and fully unfolded (right) views are CAD renders. The four labelled poses are milestones along a continuous wheel motion, during which the LiDAR sweeps through intermediate elevations.

Fig. 2: Raw depth image from Intel RealSense D435i. Black regions indicate depth voids where the IR pattern is too sparse (beyond $\sim$ 3 m) or where surface texture is insufficient for stereo matching.

Component	Prior work	Implementation here
Pixel-level RGB-D + 2D LiDAR depth fusion	[ 8 , 9 , 10 , 11 , 12 ]	Re-implemented as an upstream ROS 2 node feeding an unmodified SLAM backend.
Pseudo-3D scanning by tilting a 2D LiDAR	[ 5 , 6 , 13 ]	Driven by the existing locomotive DOF (chassis tilt) instead of a dedicated nodding servo.
Origami / flasher-pattern adaptive wheels	[ 14 , 15 , 16 , 17 ]	Re-used as both traversal mechanism and tilt actuator for the pseudo-3D scan.

TABLE I: How established techniques are combined in the prototype.

Fig. 3: Sensor fusion in the camera–LiDAR overlap sector: LiDAR returns are blended with invalid or low-confidence stereo depth pixels (confidence-gated weighted average, $w_{L}=0.7$ ), producing a more complete depth image for RTAB-Map.

Fig. 4: System pipeline: raw sensor data flows through the fusion preprocessing node (our contribution) before reaching RTAB-Map (existing library) for final map construction.

Fig. 5: Reconstructed 3D point cloud generated by RTAB-Map using the fused depth stream (cropped from the RTAB-Map GUI). Regions that appear as voids in the camera-only raw depth are filled in by the LiDAR-augmented stream, producing a continuous representation of the scene.

Panel of Figure 7 — Fig. 6: CAD-rendered views of the origami morphing wheel in its two primary configurations.

Panel of Figure 8 — Fig. 7: CAD-rendered views of four representative wheel configurations. These images are design visualizations, not photographs of physical tests. The wheels also pass through intermediate configurations during a transition.

Fig. 8: High-level modular architecture of the robotic system. The origami-inspired chassis (center) integrates three primary subsystems: a sensor-fusion-based perception module, a Nav2-based navigation stack, and a RoArm M1 manipulation arm.

Panel of Figure 10 — Fig. 9: RoArm M1 manipulation control architecture. A MoveIt action server publishes joint trajectories that are relayed through a Python bridge node to an ESP32 MicroROS velocity controller; the controller drives the ST3215 serial servos and publishes joint states back to RViz and MoveIt for closed-loop planning.

Fig. 10: Tripod setup with Intel RealSense D435i and RPLiDAR A2M12 as a handheld mapper, detachable for integration into the robotic platform.

Fig. 11: Color and depth profile from Intel RealSense D435i. Note the depth degradation in far-field regions (

>3

m) and on textureless surfaces.

Parameter	Scenario A: Indoor Corridor	Scenario B: Outdoor Sunlit
Runs per configuration	3	3
Configurations compared	Camera-only / Fused	Camera-only / Fused
Run duration	$\sim$ 120 s	$\sim$ 90 s
Depth frames per run	$\sim$ 3.6k	$\sim$ 2.7k
Depth resolution	$848\times 480$	$848\times 480$
Lighting	Fluorescent only	Direct sun + shade

TABLE II: Experimental protocol. Trajectories were manually piloted and approximately replicated across runs.

	Scenario A: Indoor Corridor			Scenario B: Outdoor Sunlit
Metric	Cam-only	Fused	Improvement	Cam-only	Fused	Improvement
Invalid depth, full frame	21%	11%	$\downarrow\,48\%$	48%	18%	$\downarrow\,63\%$
Invalid depth, stressor ROI	26%	9%	$\downarrow\,65\%$	62%	21%	$\downarrow\,66\%$
Configured depth limit	$\sim$ 3 m	8 m	$\uparrow\,5\,\mathrm{m}$	$\sim$ 3 m	8 m	$\uparrow\,5\,\mathrm{m}$
Input disagreement †	4% before fusion			27% before fusion
Loop closure, return leg	1/3 (33%)	3/3 (100%)	$\uparrow\,67\,\mathrm{pp}$	N/A	N/A	N/A

TABLE III: Depth-stream measurements in the two test scenarios (means over three runs per configuration). Green arrows mark favorable changes from the camera-only stream: lower invalid-depth fractions or higher configured depth limit and loop-closure rate.

Explore similar work

Learning Locomotion on Discrete Terrain via Minimal Proximity Sensing
Jun 30, 2026Jiale Fan, Connor Flynn, Tianao Xu +4Proprioception Legged Robots
All authors are with the Robotic Systems Lab, ETH Zurich.
UniPoint: Unified Point-Level Sensor Fusion for Humanoid Locomotion Across Challenging Terrains
Sep 20, 2026Sicen Li, Zhen Chu, Chao Li +2Event-Guided Sensor-Level Fusion Robotic Perception
College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China · Yunshenchu Technology Co., Ltd., Hangzhou, China · Zhejiang Key Laboratory of Additive Manufacturing Technology and Equipment, Hangzhou, China
PIVOT: Perception-aware Independent Viewpoint Online Optimization
Sep 16, 2026Yuyang Chen, Shekoufeh Sadeghi, Charuvahan Adhivarahan +4Robotic Perception Visual Inertial Odometry
Department of Computer Science and Engineering, University at Buffalo, Buffalo, NY 14260, USA · Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 32603, USA

Jun 30, 2026, cs.ROJ/K move · Enter open · S save

Learning Locomotion on Discrete Terrain via Minimal Proximity Sensing

Jiale Fan, Connor Flynn, Tianao Xu, Junzhe He+3

All authors are with the Robotic Systems Lab, ETH Zurich.

Proprioception Legged Robots Terrain Robotic Perception Proximity Operations

Open paper AskPDF arXiv

Learning-based control has revolutionized dynamic locomotion, yet navigating unstructured terrain remains limited by a robot's incomplete awareness of imminent ground contact. While global perception systems such as LiDARs and depth cameras provide environmental context, they are frequently plagued by latencies, occlusions, and the high computational cost of dense geometric reconstruction. On the other hand, proprioceptive feedback is purely reactive, initiating corrections only after impact has occurred. This work explores embedding a minimal suite of low-cost, high-frequency infrared proximity sensors directly into the feet of a quadrupedal robot. These sensors provide "pre-contact" feedback that is robust to self-occlusions and significantly less computationally demanding than conventional vision-based pipelines. By integrating these localized signals into a reinforcement learning framework, we enable the robot to anticipate terrain discontinuities such as gaps and stepping stones that are problematic for traditional perception stacks due to occlusions or state estimation drift. We demonstrate that such sparse, near-field sensing can be reliably modeled in simulation and transferred to the real world with high fidelity. Experimental results show that local proximity sensing substantially improves traversal robustness over discrete terrain and offers a low-power, low-latency alternative or complement to complex global perception suites in unpredictable environments. For more information about results and methods, please see the project website: https://sites.google.com/view/foot-tof/home.

Abstract

Explore similar work

Learning Locomotion on Discrete Terrain via Minimal Proximity Sensing

UniPoint: Unified Point-Level Sensor Fusion for Humanoid Locomotion Across Challenging Terrains

PIVOT: Perception-aware Independent Viewpoint Online Optimization